In a transmission of a sound data through a circuit switching network or packet network, a coding and decoding are executed to transmit and to receive a sound signal. As a sound compression method, for example, an ITU-T (International Telecommunication Union Telecommunication Standardization Sector) recommendation G.711 method and a CELP (Code-Excited Linear Prediction) method have been known.
When a sound data coded based on such a compression method is transmitted, in some case, a portion of the sound data can be lost due to an error relevant to radio communication or due to congestion of the network. As for error compensation for the lost portion, a sound signal corresponding to the lost portion is generated based on information of the preceding portion of the sound data to the lost portion.
In such error compensation, sound quality may degrade. Japanese Laid Open Patent Application (JP-P2002-268697A) discloses a method to reduce the degradation of sound quality. In the method, a filter memory value is updated by using sound frame data included in a packet received at late timing. In other words, when the packet of loss is received at late timing, the sound frame data included in the packet is used for updating the filter memory value which is used by a pitch filter or a filter representing outline of spectrum.
Japanese Laid Open Patent Application (JP-P2005-274917A) discloses art relevant to ADPCM (Adaptive Differential Pulse Code Modulation) coding. The art can solve a problem that mismatch between the states of predictors of coding side and decoding side causes unpleasant noise. The problem may occur in case that correct coded data is received after the loss of coded data. In a predetermined duration after transition of the state of packet loss from “detect” to “not detect”, a detection state controlling section gradually reduces an intensity of compensation signal generated based on sound data of the past. Since the states of the predictors gradually match and sound signal gradually become normal in the course of time, the intensity of the sound signal is permitted to increase gradually. Consequently, the art can take an effect that the unpleasant nose is not outputted even just after restoration from the loss state of coded data.
Japanese Laid Open Patent Application (JP-A-Heisei, 11-305797) discloses a method in which a linear prediction coefficient is calculated from a sound signal and a sound signal is generated based on the linear prediction coefficient.